CASIA OpenIR  > 毕业生  > 博士学位论文
基于单目视觉的运动目标参数估计
其他题名Parameters Estimation of the moving target based on the monocular vision
陈国栋
学位类型工学博士
导师徐德
2013-05-22
学位授予单位中国科学院大学
学位授予地点中国科学院自动化研究所
学位专业控制理论与控制工程
关键词打乒乓球机器人 视觉测量 姿态估计 运动特征提取 旋转信息估计 智能相机 Table Tennis Robot Visual Measurement Pose Estimation Extraction Of Movement Features Rotational Information Estimation Smart Camera
摘要在机器人应用领域,视觉引导是控制机器人移动作业的关键技术和重要途径。对于打乒乓球机器人而言,视觉系统是其实现自我感知与环境交互,并捕捉环境中感兴趣目标的重要手段。基于打乒乓球机器人系统平台,本文对单目视觉系统下的运动目标参数估计等问题展开深入的研究,主要涉及到打乒乓球仿人机器人击球过程中的本体姿态估计、运动员击球过程中球拍的运动特征轨迹提取、乒乓球旋转参数的在线估计及用于架构视觉系统的智能相机样机的研制等问题。具体来说,论文主要工作包括: 第一,提出了基于HSV空间的自适应阈值分割算法,稳定地实现了目标的图像分割;分析了图像模糊边缘的成因,提出了一种基于概率统计的亚像素精度边缘检测算法,将边缘点的定位精度细化到亚像素级,实现了稳定可靠的边缘提取;采用了基于PnP位姿初值结合OI优化的位姿估计算法,快速稳定地实现了本体姿态的估计,既保证了系统的实时性又保证了位姿估计的精度。 第二,提出了一种多约束、宽阈值的目标分割算法,解决了反光强烈、光线不足等因素的干扰问题,保证了目标分割的鲁棒性和准确性;提出了一种1-D LDO边缘线性检测算子,能够实时稳定地提取出特征线有效边缘,并实现边缘骨架的细化工作;优化了Hough变换算法,并将其与RANSAC算法和最小二乘方法结合,提高了特征线的提取精度和鲁棒性;依据球拍击球的运动事实,准确有效地提取出击球阶段的球拍运动特征轨迹。 第三,讨论了乒乓球旋转的成因,根据旋转的主要特征把旋球分为6个基础类别;分析出决定旋转模式及大小的关键因素,为旋球参数估计提供理论依据;依据运动员击球动作的先验知识,设计了包含法向连续性滤波、位移平滑滤波、位移法向角滤波、采样间隔滤波的多重滤波器,有效地剔除误判的运动特征采样点;提出了一种模糊信息融合的参数估计算法,对球拍运动特征采样点旋转参数进行模糊估计,然后利用轨迹中的有效旋转信息进行信息融合,获得最终的旋球估计参数。 第四,搭建了基于DSP+FPGA的嵌入式视觉单元的硬件平台;架构 了FPGA的内部系统框架,实现了SCCB协议并完成了OV5620的图像采集及乒乓存储操作;设计了总线占用轮询机制,实现了多模块对外部存储器的优先级访问;完成了DSP对FPGA的异步数据读取,实现了RAW-RGB到RGB转换的图像处理算法,建立了基于UDP的网络通讯,并完成部分上位机软件开发工作,实现了对智能相机的远程操作。 最后,对本文的研究成果进行总结,并指出进一步需要开展的工作。
其他摘要In the fields of robotic applications, visual guidance is a key function and an important way of controlling the robot's movements. As for the table tennis robot,the vision system is a significant means of robot's self-perception and contacting with environment, which is also be used to capture the attractive target. In this thesis, based on the robotic table tennis system, the parameters estimation of the moving target are further investigated with the monocular vision system, mainly referring to pose measurement for humanoid robot, feature extraction of racket's movements in the striking process, rotational velocity prediction of the flying ball, the development of smart cameras used in the vision system and so on. The main contributions of the thesis include the following issues: Firstly, a novel adaptive threshold segmentation algorithm in HSV color space is proposed, based on which, the target image could be segmented stably. The cause of the image's fuzzy edge is analyzed. After that, a statistics-based edge detection algorithm with sub-pixel accuracy is presented to achieve a stable and reliable edge extraction. The accuracy of edge extraction can reach to a sub-pixel level. The body posture is estimated quickly and stably by combining the initial PnP value with OI optimization, and the real-time performance of the system and the accuracy are achieved as well by using the method. Secondly, a target segmentation algorithm with multi-constraints and wide-threshold is presented, eliminating the disturbing factors such as high reflection, low light and so on, and thus the robustness and the accuracy of the target segmentation are guaranteed. Additionally, a linear 1-D LDO edge detector is proposed to stably extract the effective edge of the characteristic lines in real time, and achieve the skeleton of thinning. Meanwhile, Hough transform algorithm is optimized and combined with RANSAC algorithm and least squares method, and in this way the accuracy and robustness of the feature lines extraction are improved. According to the facts of racket's striking movement, the feature trajectory of the racket in the striking process is extracted. Thirdly, the cause of the ball's rotation is discussed, and the rotation is divided into six categories according to the main rotational features. The crucial factors that influence the rotating mode and rotating speed are investigated, which provide the theoretical basis for the estimation of spinning ball's paramete...
馆藏号XWLW1854
其他标识符201018014628002
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/6508
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
陈国栋. 基于单目视觉的运动目标参数估计[D]. 中国科学院自动化研究所. 中国科学院大学,2013.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
CASIA_20101801462800(25510KB) 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[陈国栋]的文章
百度学术
百度学术中相似的文章
[陈国栋]的文章
必应学术
必应学术中相似的文章
[陈国栋]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。